Error analysis in fixed-point digital filters using sign-magnitude truncation

The quantization error of a digital filter employing fixed-point arithmetic with sign-magnitude truncation is analyzed. The effect of coefficient quantization is also included. Exact analyses are presented first for Gaussian, sinusoidal and Gaussian plus sinusoidal inputs. Quasi-linearization is then employed to yield a simple computation method of the output noise variance. The resulting expression contains parameters that depend on the input signal statistics. An upper bound is given that is independent of these statistics. Explicit expressions are given for the second-order sections of the digital filter realized in the D1 and D2 forms together with a general expression for the cascade connection of the second-order sections.